How to forecast green energy production - The recipe for success
In the absence of suitable energy storage solutions, you must be able to continually balance electricity production and consumption. That’s one of the main challenges holding back the growth of renewable energy sources such as solar PV and wind power, with their inherent variability. It’s not just that you can’t rely on solar during the long winter nights: it only takes a cloud or a gust of wind to affect production levels!
It’s about knowing how to balance the energy mix
The variability of renewables has a direct financial impact along the whole value chain. It affects the energy producers, who are constantly at risk of either over-producing (and having to somehow get rid of the surplus so as not to overload distribution infrastructures) or under-producing (and losing potential revenues). So they have to factor that uncertainty into their financial and operational models. It affects the aggregators, who buy electricity from smaller producers in order to resell it wholesale, and play a key role in this balancing process on the networks. It affects distributors, whether they are small-scale local distributors or they manage national grids, They need to have a highly accurate and up-to-date picture of current production in order to offer a high quality service. And that’s even before you take into account so-called ‘prosumers', customers who actually generate some of their own electricity!
Being able to forecast production very accurately is crucial for the economic viability of the renewables sector, as well as for its ability to compete with other sources of energy and to increasingly dominate the energy mix amid efforts to fight climate change. The European Union, most notably, has set itself the target for renewables to provide 32% of end-user energy consumption by 2030 (compared with 18.9% in 2018).
Putting the theory into practice
Forecasting energy production is a complex process because it not only depends on meteorological conditions, but also on the nature of the infrastructure involved: the orientation and surface area of solar panels; the height of wind turbines; the length of the turbine blades… But Soljex has developed models for both wind and solar generation that take these physical characteristics into account. The models – known as SF Wind and SMF Solar – can facilitate simulations that can be used to assess a potential energy production site, for example, or to provide forecasts as soon as a new installation is commissioned, when no historical data is yet available. However, the energy produced by a wind turbine is directly proportional to the wind-speed cubed, so errors very quickly get amplified. If you’re looking for much more accurate forecasts, you have to go much further than these theoretical models will allow, as they cannot take into account all the myriad technical and operational parameters involved.
By Amr Shaheen
Chief Executive Officer
To do this, you need call on Machine Learning: by comparing historical production data with current measurements, taken moment by moment, the algorithm is able to build an extremely precise forecasting model (in statistical terms) which is unique to each particular installation.
“To meet this need,“We start from a physical model representing the characteristics of the installation, provided by the operator, and at the same time we set up the process for gathering production data which will enable us, over time, to develop a statistical model. The two models are then set to regularly ‘compete’ with each other, until the statistical model eventually wins! Then it takes over, although it constantly continues to refine itself.”
Learning to correct itself
Of course, as with any approach based around data, everything depends on the quality of that data. “Effectively, if there is a problem with one of the solar panels, for example, it’s important to take it out of the dataset so it doesn’t skew the analysis and risk disrupting the calibration of the model. That’s why the model that we have designed actually adapts itself: in other words, it is capable of relearning from cleaned data if it has detected a divergence in the trend. Today, for our next-day electricity production forecasts, we aim for an average error margin of less than 15% on solar and 10% on wind power.
“By combining high-quality weather forecasts, proven physical models and expertise in algorithms, our solution gives everyone involved in renewable energy access to forecasts that are sufficiently precise to reduce the levels of uncertainty in their business, and sufficiently timely so that they can organize themselves accordingly, for example by scheduling maintenance operations during slack periods.
A dual innovation that could prove decisive in the future growth of renewables.